Feature Extraction
Face mimics result from the dynamic interaction between 64 facial muscles
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
299
and they can be characterised through its movement pattern. The goal of the feature
extraction is to register this movement pattern and to reduce the vast data amount to
the essential values of the face surface moving.The developed feature extraction
approach consists of four following steps:feature point labelling, shortly FP-
Labelling
feature points tracking, shortly FP-Trackingfeature points normalisation ,
accordingly FP-Normalisation
feature vector generation and quantisation
MPEG-4 Facial Animation (FA) is the part of the standard, which works on the
“efficient Coding of form and animation of human faces” [2]. MPEG-4 defines 84
Feature Points (FPs) in several regions of a neutral face. These points fulfil two tasks:
on the one hand through them the form of the face can be defined and on the other
hand they work as reference points of the Facial Animation parameters. Fig. 2a shows
the, in the MPEG-4 Standard defined, Feature Points [2]. Facial Animation Parameter
(FAPs) of the MPEG4 correspond to minimal recognisable muscle movements and
are comparable with the by Ekman and Friesen introduced FACS Action Units [3].
Through our tracking process 19 points from 4 facial regions: eye brows, eyes,
nose, mouth - are defined. Firstly they are labeled manually in every first image of a
sequence and then they are tracked using the Lucas-Kanade-algorithm over all
frames. There are 3 stable und 16 dynamic Feature Points, which will be extracted
from these regions (Fig. 2b). The stable Feature Points are only used for the FP-
Normalisation, while the dynamic Feature Points are used for the feature vector
generation. Stable FPs define points, that are not movable by facial muscles, so they
are suitable to recognize a movement of the head and if needed to correct this
movement (FP 3.8, 3.11 and 9.15 in Fig 2a). Dynamic FPs are points, which the
human being moves with his mimic, and they are suitable for the feature vector
generation (FPs 4.6, 4.4, 4.2, 4.1, 4.3, 4.5, 3.2, 3.4, 3.1, 3.3, 9.2, 9.1, 8.4, 8.1, 8.3 and
8.2 in Fig. 2a).
Figure 2: a) MPEG4 Facial Animation, Feature Points, b) FP-Labelling
After manual labelled FPs are tracked automatically from frame to frame. This
approach is called Feature Point Tracking (FP-Tracking), attempting to follow the
movements of FPs. The implemented FP-Tracking is based on the Lucas-Kanade-
algorithm [4]. The detailed report of the implemented method is described in [5]. A
detailed examination of the widely spread method derivatives by Lukas-Kanade was
taken through in the framework of this project in [6].
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
300
Virtual animation
All detected and extracted parameters of face mimic tracking are registered in a
Face Mimic Graph. The virtual simulation of the facial expression combines the 3D-
modeling of the virtual face and the transferring of the Face Mimic Graph to it. The
tracked animation of the real face will be cloned on the virtual face using the MPEG4
facial animation coding.
Figure 3 - FP-tracking results
Figure 4 - Virtual Face animation using face mimic tracking
5. Project: Vision-based driver assistance system
The prototype Vision-based driver assistance system (VDAS) [1,2] is
developed in Kooperation with Infineon Munich. An automatic recognition of traffic
signs could inform the driver about actual speed limits or overtake restrictions, which
could be ignored in difficult traffic situations. For that purpose an efficient detection
of traffic sign candidates (denoted as traffic sign detection – TSD) and a successive
reliable classification of these candidates are required.
The prototype system is implemented on a new multi-core processor (VIP-II),
developed by the Infineon Technologies AG [1]. The multi-core architecture supports
data-parallel and task-parallel processing. A computational power of 14 GIPS could
be used for data-parallel operations, while the computational power for control-based
operations is similar to a 1.2 GHz processor. Power consumption for the current
prototype is less than 500mW.
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
301
Virtual Simulation of traffic scenes
The traffic sign detection is based on a trainable classifier-cascade as proposed
by Viola and Jones. The system requires a large set of positive and negative training
data (several thousand images). As the final system has to handle different weather
and illumination conditions as well as combinations of rotation, scale, occluding,
pollution etc. the training data has to include all these variations in a sufficient
number of samples. Obviously, the generation of this training set from “real” images
is time consuming and sometimes it is even impossible to capture all required traffic
sign variations. Therefore, we use computer-graphics generated training images
instead of “real” images. A 3D simulation of traffic scenes based on the modelling of
all necessary objects, illuminations and weather conditions allows a parameterized
rendering of the virtual images. The parameterized rendering process could also
include the simulation of the projection behaviour of a real camera. This approach
allows the generation of a sufficient set of training data even under extreme weather
or illumination conditions. The artificial trainings data are currently used for vehicle
detection, detection and tracking of road lanes as well as for the classification of
traffic signs.
a
b
Figure 5 - Comparison of real camera image (a)
and a computer- graphics generated image (b)
Automatic detection and classification of traffic signs
The detection of traffic signs in 2D images is based on a cascade of trainable
classifiers introduced by Viola and Jones. To detect signs inside an image the search
window is scanned over the image. At every position of the search path, the image is
analyzed through a cascade of classifiers. If one of the classifiers fails, the analysis
process is stopped and the search window is moved to the next position. A positive
classification of a traffic sign is only possible, if all cascades return a positive result.
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
302
This approach is time efficient because negative patterns could be discarded early
without the computation of all required features. To detect traffic signs at different
scales, the search window inclusive the set of trained Haar-like features is scaled over
a range of defined steps. The classifier-cascade consists of 18 single classifiers and
overall about 200 Haar-like features are used. Together this requires 15kBytes for
memory.
The assignment of traffic sign candidates, which are generated by the previous
traffic sign detection step, to the correct sign type requires a set of specific features as
input for the neural network classifier. For classification a multilayer perceptron net
with a feed forward topology is used. To distinguish between all types of traffic
signs, two separate networks are used. The input layer of the neural net contains 256
neurons. The number of output neurons corresponds to the number of signs, which
have to be distinguished, plus one additional neuron to indicate the negative class
(therefore 15 neurons are used for speed limit signs and 16 neurons for the
abolishment class). The final traffic sign recognition system is executed on a
dedicated multi-core processor, which is based on fix-point arithmetic. Therefore, the
neural network implementation supports fix-point arithmetic. The neural network is
executed in several parallel threads to achieve the maximal performance.
For the verification of the proposed traffic sign detection approach, real video
sequences were recorded. The camera was mounted in a passenger car and runs with
25 frames per second. Over 150 different signs were captured. Due to the camera
motion each sign is visible over 5-25 frames. A total of 2000 images with traffic
signs were available, 1400 images contain traffic signs in the resolution range from
16x16 to 30x30 pixels. The rate of correct detected traffic signs is about 89%. The
false positive rate is 0,027 wrong detected traffic sign candidates per frame. The
average positive classification rate is about 97% for the network that was trained with
“real” test data and about 73% for the network that was trained only with artificial
data. The whole system takes 40ms on the multi-core processor, where 10 worker
threads consume 62.5% of the available performance. This means that 25 frames can
be handled each second.
Figure 6 - Examples of detected traffic sign candidates.
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
303
Figure 7 - Detection & classification results for real camera images
Lane Departure Warning
Beside the traffic sign detection a lane departure warning system was
implemented. The approach uses several search lines to find the edges of the road
markings. Based on these measure points a linear lane model is derived. In the next
step we use an extended Kalman-Filter and a clothoid model to follow the road and
become a appropriate prediction. The LDW-system is executed in several parallel
threads to achieve the maximal performance. Therefore the implementation fulfills
the real-time constrain.
Real-time Detection of Traffic Signs on a Multi-Core Processor, R. Ach, N.
Luth, A. Techmer, accepted for IEEE Intelligent Vehicles Symposium, 2008, 4-6
June 2008
Classification of Traffic Signs in Real-Time on a Multi-Core Processor, R.
Ach, N. Luth, T. Schinner, A. Techmer, S. Walther, accepted for IEEE Intelligent
Vehicles Symposium, 2008, 4-6 June 2008.
СЕТЕВЫЕ ТЕХНОЛОГИИ ПРОГРАММИРОВАНИЯ –
ЭТО БУДУЩЕЕ ВЕБ-РАЗРАБОТКИ
А.Г. Ни
Казахский национальный технический университет им. К.И. Сатпаеа
With growth of popularity of network Internet more and more involve attention
network technologies of programming.
In this connection in the present article the curriculum maintenance
(SYLLABUS), developed for a course «Network technologies of programming» for
students of chair «the Software of systems and networks» by KazNtU of K.I.Satpaeva
of a speciality 050602 "Computer science" is considered
В связи с возрастающей популярностью сети Internet все более
привлекают внимания сетевые технологии программирования, которые
включают в себя языки программирования такие, как Java, Perl, PHP, языки
описания документов – HTML и XML.
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
304
Сетевые технологии программирования - это то, что позволяет создавать
сайты, заставляет модули работать, почту отправляться, обмениваться данными
и многое другое.
Однако с развитием Internet-технологий все более актуальной становится
проблема связанная с хранением структурированных данных (взамен
существующих файлов баз данных) и обеспечением совместимости при
передаче структурированных данных между разными системами обработки
данных, особенно при передаче таких данных через Интернет;
Наиболее удобный механизм, позволяющий свести к минимуму
проблемы внутрифирменных форматов данных это - обмен информацией в
формате XML.
XML (Extensible Markup Language) - это язык разметки, описывающий
целый класс объектов данных, называемых XML-документами. Этот язык
используется в качестве средства для описания грамматики других языков и
контроля над правильностью составления документов, то есть он представляет
собой метаязык, являющимся основой для создания новых языков. Очевидно,
что за XML- технологией большое будущее.
В связи с вышесказанным в настоящей статье рассматривается
содержание учебной программы (SYLLABUS), разработанной для курса
«Сетевые
технологии
программирования»
для
студентов
кафедры
«Программное обеспечение систем и сетей» КазНТУ им. К.И.Сатпаева
специальности 050602 «Информатика».
Содержание программы включает в себя следующие темы:
Сетевые технологии программирования. Основные понятия и подходы.
Инструментальные средства создания XML- документа.
Корректно сформированные XML-документы.
Валидные XML-документы. Специализированный язык DTD.
Валидные XML-документы Специализированный язык XML-схема.
Отображение XML-документов с использованием таблиц каскадных
стилей.
Отображение XML-документов с использованием связывания данных.
Отображение XML-документов с использованием XSL-таблиц стилей
XML и технологии баз данных.
Изучение курса «Сетевые технологии программирования» включает в
себя чтение лекций, проведение лабораторных и самостоятельных работ.
Разработанная программа (SYLLABUS) отводит на курс 3 кредита, из
них 2 кредита на лекции и 1 кредит на лабораторные работы.
В программе предусмотрено выполнение студентами следующих
лабораторных работ:
Отобразить базу данных в HTML – документе, согласно варианту задания
Создание XSL - файла для отображения содержимого XML – документа.
Использование внутреннего DTD в XML –документе. Создание
отдельного файла с DTD и подключение его к XML – документу.
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
305
Использование внутренней XML-схемы в XML –документе. Создание
отдельного файла с XML-схемой и подключение его к XML – документу.
Отображение созданного XML – документа с помощью HTML.
Создание каскадной таблицы стилей в отдельном файле для отображения
содержимого XML – документа.
Отобразить базу данных в XML – документе, согласно варианту задания
Также предполагается выполнение студентами самостоятельных работ в
виде реферата:
Подготовка реферата по теме «Языки описания ссылок XML - XLink,
XPointer»
Подготовка реферата по теме «Язык адресации XML - XPath
Подготовка реферата по теме «Язык описания веб-сервисов – WSDL»
Подготовка реферата по теме «Протокол передачи данных – SOAP
Подготовка реферата по теме «Язык математических формул – MathML
Подготовка реферата по теме «языки преобразования документов XML -
XSLT
Подготовка реферата по теме «Словарь, описывающий формат книги -
FB2»
Подготовка реферата по теме «объекты форматирования языка таблиц
стилей для XML- XSL-FO
В результате изучения дисциплины студенты должны:
иметь представление
о XML- технологиях;
знать
основные правила и приемы создания XML-документов на
основе официальной спецификации W3C (World Wide Web
Consortium);
уметь
создавать XML-документы с использованием различных XML-
словарей.
владеть навыками
работы с различными XML-редакторами.
Литература
Александр Печерский. Язык XML – практичкское введение. Часть 1 и 2. –
http://www.citforum.ru/internet/xml2/index.shtml
USING OF ICT FOR ENVIRONMENTAL EDUCATION
Galiya Nurmukhanbetova CandSc (Biology)
IITU
The article considers the ICT potential for increasing access to and improving
the relevance and quality of education. The potential of ICT to promote the
acquisition of obtained skills is tied to its use a tool for raising educational quality,
including promoting the shift to a learner-centered environment. Carrying out of ICT-
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
306
based tutorials, particularly using of several web sites on energy saving and the
feedback obtained from the play games will help students to change lifestyle in
sustainable pattern.
One defining feature of ICT is their ability to transcend time and space. ICT
make possible asynchronous learning, or learning characterized by a time lag
between the deliver of instruction and its reception by learners. Online courses
materials, for example, may be accessed 24 hours a day, 7 days a week. ICT-based
educational delivery also dispenses with the needs for all learners and the instructor
to be in one physical location. Additionally, certain types of ICT, such as
teleconferencing technologies, enable instruction to be received simultaneously by
multiple, geographically dispersed learners (i.e., synchronous learning).
Teachers and learners no longer have rely solely on printed books and other
materials in physical media housed in libraries (available in limited quantities and too
expensive to be purchased) for their educational needs. With Internet and the World
Wide Web, a wealth of learning materials in almost every subject and in a variety of
media can now be accessed from anywhere at anytime of the day and by an unlimited
number of people. This is particularly significant for many universities that have
limited and outdated library resources, including teaching materials in English. ICT
also facilitate access to resource persons – mentors, experts, researchers,
professionals, business leaders, and peers – all over the world.
One of the most commonly reasons for using ICT in the classroom has been to
better prepare the current generation of students for a workplace where ICT,
particularly computers, the Internet and related technologies, are becoming more and
more ubiquitous. Technological literacy, or the ability to use ICT effectively and
efficiently is thus seen as representing a competitive edge in an increasingly
globalizing job market. Technological literacy, however, is not the only skill well-
paying jobs in the new global economy will require. It has indentified what it calls
“21
st
Century Skills”, which includes digital age literacy (consisting of functional,
visual, scientific, technological, information, cultural, a global awareness) inventive
thinking, higher-order thinking and sound reasoning, effective communication, and
high productivity (Table 1).
Table 1 Skills needed in the workplace of the future
Digital Age Literacy
Functional literacy
Ability to decipher meaning and express ideas in a range of media; this
includes the use of
images, graphics, video, charts and graphs or visual literacy
Scientific literacy
Understanding of both the theoretical and applied aspects of science
and mathematics
Technological literacy
Competence in the use of information and communication
technologies
Information literacy
Ability to find, evaluate and make appropriate use of information,
including via the use of ICT
Cultural literacy
Appreciation of the diversity of cultures
Global awareness
Understanding of how nations, corporations, and communities all over
Жоғары оқу орындарында ақпараттық технологияларды оқыту сапасын жақсарту:
жолдары мен мүмкіндіктері
307
the world are interrelated
Inventive Thinking
Adaptability
Ability to adapt and manage in a complex, interdependent world
Curiosity
Desire to know
Creativity
Ability to use imagination to create new things
Risk-taking
Ability to take risks
Достарыңызбен бөлісу: |